Effects of word frequency on semantic memory in ... Language Issue...Condray et al. N400 in...

Condray et al. N400 in Schizophrenia

Effects of word frequency on semantic memory in schizophrenia:

Electrophysiological evidence for a deficit in linguistic access

Ruth Condray 1, Greg J. Siegle 1, Matcheri S. Keshavan 1,3 , Stuart R. Steinhauer 1,2

1 Western Psychiatric Institute & Clinic

Department of Psychiatry

University of Pittsburgh School of Medicine

3811 O’Hara Street

Pittsburgh, PA 15213 2 VA Pittsburgh Healthcare System

7180 Highland Drive

Pittsburgh, PA 15206 3 Department of Psychiatry

Beth Israel Deaconess Medical Center

Harvard Medical School

Boston, MA

* Corresponding author: Ruth Condray, Ph.D., Western Psychiatric Institute & Clinic,

Department of Psychiatry, University of Pittsburgh School of Medicine, 3811 O’Hara Street,

Pittsburgh, PA 15213; email: [email protected]

File name: Condray etal N400_Sz Semantic priming-WF - final 101009.doc 1

mailto:[email protected]


Abstract

Background: Abnormal storage and/or access are among the hypothesized causes of semantic

memory deficit in schizophrenia. Neuropsychological and connectionist models have emphasized

functional systems that serve the processing of word meaning and frequency: semantic storage

disturbance is presumed to result from weak representations of word meaning; defective access is

assumed to result from compromises to pathways that activate word frequency knowledge. Candidate

biological systems include neuromodulatory pathways that normally function to enhance neural signals

(e.g., cholinergic system). Electrophysiological responding may be informative regarding the storage-

access distinction for schizophrenia. Methods: Visual event-related potentials were recorded for 14

schizophrenia outpatients receiving atypical antipsychotics, and 14 healthy controls group-matched to

patients on age, gender, and demographics. N400 was elicited using an incidental semantic priming

paradigm, in which semantic relatedness and word frequency were varied, and a letter probe task.

Results: Compared to controls, patients showed reduced N400 (µV) discrimination of semantic

relatedness. Groups also showed different patterns of N400 to word frequency. Controls’ N400

increased in negativity as words decreased in frequency of occurrence, while patients did not show a

linear relationship between N400 and word frequency. Groups also differed for N400 to frequently

occurring words. Patients exhibited increased negativity to high and very high frequency words,

compared to controls. A subgroup of patients receiving antipsychotics with known affinity binding for

muscarinic receptors (clozapine, olanzapine) showed significant albeit limited N400 priming, but their

N400 to word frequency remained nonsignificant. Conclusions: Results suggest a deficit in semantic

access for schizophrenia, as well as an influence of neuromodulators on the activation of connections

among semantic representations. Cumulative findings indicating only limited N400 priming for

patients receiving either typical or atypical antipsychotics support the hypothesis that semantic

memory deficit represents a trait marker for schizophrenia.

Keywords: N400, semantic relatedness, word frequency, schizophrenia, atypical antipsychotics



Introduction

Memory impairment is an associated feature of schizophrenia with deficits observed across a

broad range of memory systems and processes, including disturbance to semantic memory (reviews:

Aleman et al., 1999; Condray 2005; Kuperberg et al., in press; Minzenberg et al. 2002). Semantic

memory represents a person’s cumulative knowledge about the world (e.g., name of the first president

of the United States, defining features of tables, etc.) (Tulving 1972; Schachter et al., 2000), and, in

this respect, provides a dynamic record of a person’s ongoing learning and experience. Language

serves as an important interface between the environment, learning, and memory, and is regarded by

some theorists as a key cognitive input system (Fodor 1983). Language impairment involving deficits

in listening and reading may predate the onset of schizophrenia (Cannon et al. 2002; Crow et al. 1995),

and is observed in patients across clinical states and medication regimens, and in their healthy family

members (reviews: Condray 2005; DeLisi 2001; Kremen et al., 1994; Minzenberg et al., 2002).

Storage and activation of linguistic representations within semantic memory are therefore an important

focus for theories of cognition in schizophrenia.

Nature of the semantic memory deficit for schizophrenia: Storage versus Access? A distinction

between semantic storage and access has long been emphasized in cognitive psychology and

neuropsychology, and, more recently, in computational neuroscience, and this emphasis could be

informative for our efforts to understand semantic memory in schizophrenia. Elizabeth Warrington

and her colleagues early dissociated word comprehension deficit due to deficient memory access

(access/refractory impairment) and deficit caused by a degradation of stored memorial representations

(degraded-storage deficit). This contrast evolved from their assessments of patients who presented to

neurology clinics with impairment in word comprehension (Warrington 1975; Warrington and Shallice

1979; Warrington and Cipolotti 1996). Their examinations of patient performance profiles suggested

that word comprehension deficit differs as a function of the type of neurological disorder. A storage

deficit involving cell-based degradation of semantic representations was proposed for degenerative

conditions, such as Alzheimer’s dementia, Pick’s disease, and viral infections. In contrast, vascular

etiologies were thought to underlie access dysphasias characterized by only sporadic availability of

semantic representations. Based on their cumulative data, Warrington et al. emphasized four factors

that distinguish the two profiles: semantic relatedness (priming); word frequency; presentation rate;

response consistency (trial-by-trial variability for identical words). A deficit in the access of semantic

memory was hypothesized to involve reduced memory accuracy for words presented at rapid rates, no



accuracy advantage for high frequency words, an atypical accuracy advantage for semantically

unassociated words (atypical or negative priming), and inconsistent accuracy across trials and time

points. In contrast, a disorder of semantic storage was proposed to involve a primary disturbance of

knowledge about semantic relationships, with no accuracy advantage for either associated or

unassociated words (absent priming), but with preserved word frequency effects and response

consistency. These contrasting performance profiles as proposed by Warrington and Cipolotti (1996)

are summarized in Table 1.

The connectionist model developed by Gotts and Plaut (2002) extended the access/refractory–

degraded/store distinction by simulating damage to mechanisms that may account for these deficits;

damage to cells that encode and connect semantic memory representations, and damage to slow-acting

neuromodulatory pathways that normally function to enhance neural signals otherwise attenuated by

refractory processes, including disruptions to the influence of acetylcholine and norepinephrine. By

“damaging” neuromodulation and sparing stored connections and, in turn, by damaging stored

connections while sparing neuromodulation, their model produced performance that was consistent

with the data reported by Warrington and Cipolotti (1996) for the access-refractory and degraded-store

profiles, respectively. An important aspect of the Gotts-Plaut model for semantic memory in

schizophrenia is its emphasis on the neuromodulatory influence of acetylcholine on transmitter release

at pre-synaptic cells and the subsequent activation and sensitivity to excitatory input at post-synaptic

cells. Acetylcholine is known to influence learning and memory (reviews: Everitt and Robbins 1997;

Hasselmo 2006), with muscarinic acetylcholine receptors playing an important role in its effects on

cortical function and cognition (Atri et al., 2004; Broks et al., 1988; Ellis et al., 2006). Although the

role of muscarinic receptors in the pathophysiology of schizophrenia appears complex (reviewed by

Raedler et al., 2007), certain findings indicate this receptor class may be important for schizophrenia-

related cognition. Reduced muscarinic receptor density and binding have been observed for

schizophrenia patients in brain regions that serve memory and attention, including hippocampal

formation, superior temporal gyrus, and dorsolateral prefrontal and anterior cingulate cortices (Crook

et al., 2000, 2001; Dean et al., 2002; Deng and Huang 2005; Raedler et al., 2003b; Zavitsanou et al.,

2004). Moreover, a negative association has been observed between postmortem choline

acetyltransferase activity and (antemortem) cognitive function in schizophrenia patients (Powchik et

al., 1998). Because administration of anticholinergic agents is associated with reduced accuracy of

memory and cognitive performance in schizophrenia patients (Strauss et al., 1990; Sweeney et al.,



1991), and because antipsychotic medications vary in their anticholinergic load (Minzenberg et al.,

2004), it is important to consider semantic memory response within the context of their mechanisms of

action. A characteristic of some atypical antipsychotic agents, such as clozapine and olanzapine, is

their affinity for muscarinic acetylcholine receptors (Chew et al., 2006; Raedler et al., 2000, 2003a).

And while atypical antipsychotics may be generally superior to typical antipsychotics, both for overall

cognition as well as for specific domains (attention, learning, verbal fluency, and processing speed)

(see meta-analyses by Woodward et al., 2005, 2007), their influence on electrophysiological indices of

semantic memory function, such as the N400, has not been determined for schizophrenia. Given the

well-established relationship between acetylcholine and memory function, examination of the effect of

atypical agents known to have affinity for muscarinic receptors may inform our theories of semantic

memory in schizophrenia.

Dissociation of structure and process: Recent efforts to clarify semantic memory impairment in

schizophrenia as either a deficit of storage or a disturbance of access provide a mixed picture. Based

on behavior response accuracy and latency during performance of a semantic priming-lexical decision

task, Rossell and David (2006) observed exaggerated word frequency and semantic relatedness

(hyperpriming) effects for schizophrenia patients, as well as response consistency across test sessions.

As reflected in Table 1, this pattern includes aspects of both storage and access deficits as defined by

Warrington and Cipolotti (1996). Laws et al. (2000) used picture naming of line drawings, which

portrayed a range of commonly- to rarely-occurring objects, to evaluate semantic memory in chronic

schizophrenia patients. Group performance accuracy was consistent across test sessions and

influenced by object familiarity (word frequency), and additional analyses of individual patient’s

performance suggested the majority were characterized by these characteristics of a storage deficit. A

few investigators have evaluated the storage disorder hypothesis by comparing semantic memory

performance between schizophrenia patients and patients with Alzheimer’s disease, which is assumed

to involve a primary disturbance of semantic storage. McKay and coworkers (1996) administered a

battery used in dementia research (Semantic Memory Test: Hodges et al., 1992) to groups of

schizophrenia patients that differed in clinical severity and age. Patients’ performance declined with

increasing clinical severity and age, and the performance of elderly schizophrenia patients was

comparable to that of Alzheimer’s patients. More recently, Doughty and colleagues (2008) selected

schizophrenia patients who evidenced cognitive impairment and compared their profile of semantic

memory deficits with the profile of Alzheimer’s patients. Performance accuracy of the two groups



differed for semantic memory reflected in naming, word-picture matching, and category sorting tasks

(Alzheimer’s < Schizophrenia), and a greater proportion of Alzheimer’s patients exhibited a significant

word frequency effect compared to patients with schizophrenia (46% versus 14%, respectively).

Doughty et al. concluded the profile of deficits observed for schizophrenia patients did not conform to

a typical disorder of semantic storage. Thus, the literature to date suggests schizophrenia may be

associated with a primary deficit in semantic storage, although a complete dissociation of structure and

process may not characterize this cognitive impairment fully.

Measurement. The type and time window of response measure, which likely capture various

aspects of the processing stream, may be important factors in tests of the storage-access distinction. To

date, the focus of schizophrenia studies addressing the storage-access distinction has involved

behavioral data, and it is therefore unknown whether the physiological responding that occurs prior to

behavioral response reflects the same admixture of storage and access disturbance described above.

For example, although Alzheimer’s disease is generally assumed to involve a primary disturbance of

semantic storage (Crutch and Warrington 2006; Hodges et al., 1992; Salmon et al., 1999; Warrington

and Cipolotti 1996), the cumulative findings show this population is also characterized by disruptions

to semantic access (early work reviewed in Nebes 1989; Rogers and Friedman 2008). Results from

event-related potential (ERP) studies underscore this point. Ford et al. (2001) reported that patients

with Alzheimer’s disease, who showed anomia involving reduced naming accuracy, nevertheless

produced a typical or normal N400 differentiation of semantic context during a picture-word

verification task. A similar finding was reported by Schwartz et al. (2003) who used semantically

congruous and incongruous sentence endings to elicit the ERP. In the Schwartz et al. study,

Alzheimer’s patients also showed the typical ERP semantic incongruity effect, although they differed

from controls with respect to temporal course. Of interest, the scalp distribution of these ERP effects

was found to differ between Alzheimer’s patients and controls in both the Ford et al. and Schwartz et

al. studies.

Neuromodulatory processes and antipsychotic medication. The effect of neuromodulation on

transmitter release may provide an account, in part, for the mixed disorder pattern observed for

schizophrenia. As noted above, antipsychotic drugs commonly administered to schizophrenia patients

vary in their pharmacodynamic profiles, with the primary action of typical antipsychotics, such as

haloperidol, believed to involve antagonism of dopamine (DA) D2 receptors in mesolimbic and

mesostriatal regions. In contrast, the action of atypical antipsychotics is thought to include a



combination of muscarinic receptor antagonism, 5-hydroxytryptamine (5-HT)2A/DA receptor

antagonism, and antagonism of extrastriatal D2 receptors in limbic and thalamic regions (Meltzer

2004). It has been suggested that the increased DA and acetylcholine release in prefrontal cortex

observed for atypical antipsychotics may account for their advantageous effects on cognition (Meltzer

2004). Medication regimens for schizophrenia patients in the studies of the storage-access distinction

discussed above involved a variety of pharmacological regimens. The Rossell and David (2006) study

included patients receiving typical (28%) and atypical (66%) antipsychotics, as well as patients who

were medication free (6%). The Doughty et al. study (2006) also included schizophrenia patients who

were receiving typical (5%) and atypical (80%) antipsychotics, as well as patients who were

medication free (10%). An additional patient was receiving lithium. The studies reported by Laws et

al. and McKay et al. did not include medication information.

Study Hypotheses: The present study examined two aspects of the storage-access distinction,

semantic relatedness and word frequency, for a sample of schizophrenia patients and age- and

demographically-matched normal controls. All patients were receiving atypical antipsychotic

medications at the time of ERP testing. The N400 component was the response measure of theoretical

interest. The study design involved an incidental semantic priming paradigm in which semantic

relatedness and word frequency were varied, and included a behavioral task (letter probe verification)

that was resource demanding, independent of the psycholinguistic factors of interest, and biased

toward a shallow level of processing (orthography). It is therefore assumed that the N400 priming

effects elicited by this paradigm reflect primarily automatic processing. Our general hypothesis was

that patients would exhibit disturbance, indexed by N400, to both storage and access of semantic

memory. Firstly, based on the prior behavioral studies of the storage-access distinction in

schizophrenia, as well as the extensive literature concerning N400 priming in this patient population,

we predicted that patients would show a deficit in semantic memory storage, which would be reflected

in a reduced and/or absent N400 semantic priming effect. Secondly, contrary to findings from prior

behavioral studies, we predicted patients would show abnormal semantic access reflected in a reduced

or absent N400 word frequency effect. Finally, results from neuropsychological studies of

schizophrenia patients receiving atypical antipsychotics would lead to the expectation that semantic

memory will benefit from atypical agents. Based on prior work (Condray et al., 1999; 2003), however,

we hypothesized that semantic memory, indexed by N400, represents a risk marker for schizophrenia

(endophenotype: Braff et al., 2007; Gottesman and Gould 2003; mediating vulnerability trait:



Nuechterlein et al., 1992), and would therefore remain significantly compromised even in patients

receiving atypical antipsychotic drugs. Secondary analyses were conducted to address the possibility

that neuromodulation due to acetylcholine may influence semantic access (Gotts and Plaut, 2002).

This possibility was evaluated indirectly by restricting these analyses to a subgroup of patients

receiving atypical antipsychotics known for their affinity binding for muscarinic acetylcholine

receptors (clozapine and olanzapine).



Materials and Methods

Sample

All study participants were evaluated with the Structured Clinical Interview for DSM-IV (First et

al. 1995-1996), and psychiatric diagnoses (American Psychiatric Association 1994) were assigned

during case conferences. General study inclusion requirements were: American English was the first

language learned, no history of major medical or neurological disorders, and vision corrected to 20/20.

Schizophrenia patients were excluded if they met DSM-IV criteria for current substance use disorder.

Participants were selected from the database of a larger family study based on the presence of

sufficient numbers of artifact-free event-related potential (ERP) trials during the semantic priming

study (see below), and patients were further selected based on their medication regimen at ERP testing

(atypical antipsychotics). Table 2 presents the characteristics of the sample. Patient and control

groups (50% female) did not differ with respect to age, education, spelling ability, short-term memory

span, parental socio-economic status, handedness, or race (all p-values = n.s.). Groups differed for

single-word reading (WRAT-3: F 1,26 = 7.59, p = .01), verbal intelligence and vocabulary (WASI VIQ:

F 1,26 = 17.77, p < .001; Vocabulary: F 1,26 = 17.03, p < .001), and individual socio-economic status (F

1,26 = 17.42, p < .001).

Patients with schizophrenia disorder (n = 14) were recruited from Western Psychiatric Institute

and Clinic outpatient programs; 13 were diagnosed with schizophrenia and 1 was diagnosed with

schizoaffective disorder. For patients at ERP testing, the mean dosage of atypical antipsychotics

(chlorpromazine equivalents: Woods 2003) was 311.9 mg daily (median = 300; sd = ± 103.45; range:

150 – 450), with the number of patients receiving each specific drug as follows: risperidone (n= 4),

olanzapine (n= 7), clozapine (n= 1), quetiapine (n= 1). One patient was receiving both ziprasidone and

clozapine. One patient was also receiving benztropine mesylate (1 mg daily). Some patients were

receiving additional medications at testing, as follows: mood stabilizers (n= 4), antidepressants (n= 9),

and anti-anxiety medications (n=2). Normal controls (n=14) were diagnosed as having no lifetime

DSM-IV Axis I or II disorders. Only one normal control reported taking medication within the 24

hours before ERP testing (levonorgestrel and ethinyl estradiol). Each study participant was paid $140

for participation. Following explanation of study procedures and before testing, all participants

provided written, informed consent to participate using University of Pittsburgh Institutional Review

Board approved consent forms.



Study Design

The repeated measures study design involved an incidental semantic priming paradigm, with

presentation of word pairs that varied in semantic association and word frequency followed by the

behavioral task stimuli (letter probes), which were unrelated to the psycholinguistic features of the

word pairs. Prime – target word pairs (n=360) were created to represent four levels of word frequency

(very low, low, high, and very high) and two levels of semantic association (associated and

unassociated), which resulted in eight combinations of word frequency and semantic association. Each

of the eight word-pair combinations was represented by 45 trials (e.g., very low word frequency-

semantically associated = 45 trials; very low word frequency-semantically unassociated = 45 trials;

very high word frequency-semantically associated = 45 trials; etc.).

Stimuli

Word Frequency. Word pairs were created to represent a wide range of word frequencies as

reflected in word frequency counts. Prime and target words within each pair were matched for word

frequency (e.g., low frequency target paired with low frequency prime). Each word was used only

once. Prime words were nouns selected to represent the four word frequency intervals used in the

Postman (1970) study of discrete or single-word association. These frequency intervals were based on

the Thorndike – Lorge (1944) Magazine corpus (L-column) involving 4.5 million words, and

represented half-step increases on a logarithmic scale: very low = 1-3 occurrences /4.5 million

possible occurrences; low = 10 – 33; high = 100 – 333; very high = 1000 – 3333. Researchers

commonly report word frequencies based on the Brown Corpus, and descriptive statistics are also

provided based on the Francis – Kucera (1982) count of this corpus involving 1 million occurrences.

Median frequencies of prime words for each frequency interval based on the Thorndike-Lorge count

are: very low = 2; low = 19.5; high = 180; very high = 1591. Target words were also selected from

the same word frequency intervals, and the median frequencies for each frequency interval based on

the Thorndike-Lorge count are: very low = 2; low = 16; high = 175; very high = 1172.5. Median

frequencies of prime words for each frequency interval based on the Francis-Kucera count are: very

low = 1; low = 3; high = 32.5; very high = 289.5. The median frequencies of target words for each

frequency interval based the Francis-Kucera count are: very low = 1.5; low = 7; high = 31.5; very high

= 293. For associated target words, the following exceptions to these frequency intervals were

imposed by the nature of word association norms (see Semantic Context below): very low frequency

(13% = 0 occurrences/4.5 million possible occurrences; 15% = 7 – 15 occurrences); low frequency



(4% = 0 – 1 occurrences; 22% = 6 – 9 occurrences; 1 word = 55 occurrences); very high frequency (>

500 occurrences).

Word frequency (4 levels) of word pair stimuli was tested for each of the experimental

combinations of prime-target (2 levels) and semantic association (2 levels). ANOVA (word frequency

x prime-target x semantic association) showed the word frequency factor (natural log transformation)

was highly significant (F3,111 = 5534.12, p < .001, linear trend). In contrast, word frequency did not

differ between primes and targets (p = n.s.), associated and unassociated word pairs (p = n.s.), or any of

the combinations of these factors (no significant interactions involving word frequency, prime-target,

and semantic association: all p-values = n.s.).

Semantic Context. Semantically associated word pairs (n=180) were created using published

word association norms (Battig and Montague 1969; Jenkins 1970; Marshall and Cofer 1970; Nelson,

McEvoy, and Schreiber 1998; Postman 1970) and the considerations detailed above regarding word

frequency. An equivalent number of semantically unassociated word pairs (n=180) were created by

randomly ordering additional words selected on the basis of word frequency with the restriction that

the resulting pairs were not semantically related. Two strategies were used to verify degree of

semantic association for word pairs: (1) strength of association for semantically associated pairs based

on tests of discrete or single-word association reported in published sources; (2) independent ratings of

the degree of semantic relatedness for each of the 360 word pairs used in the present study.

(1) Strength of association from published sources: Most of the associated word pairs (69%) were

characterized by a direct or forward associative relationship, as determined in tests of discrete or

single-word association (“provide the first word that you think of”: Jenkins 1970; Marshall and Cofer

1970; Nelson, McEvoy and Schreiber 1998; Postman 1970). Mean associative strength for these

discrete association pairs was .113 (median/sd = .047/.162; range: .001 - .80). The remainder of

associated word pairs (31%) included 31 pairs characterized by overlapping associative relationships

most of which involved associates to shared semantic categories, such as beaver – skunk within the

category of ‘four-footed animal’ (from Battig and Montague 1969). These overlapping pairs showed a

mean associative strength of .02 (median/sd = .002/.066), computed in the manner described by Nelson

et al. (1998). Very low and low frequency words are underrepresented as discrete free-association

responses in published word association norms, which necessitated use of lexicon source materials

(Thesaurus synonyms and related words) for creating the remaining 24 associated word pairs.



(2) Semantic relatedness ratings: Independent ratings of prime-target word relatedness were

recorded for each of the 360 word pairs used in the present study. Ratings were made by a subset of

the study sample (n=17; 53% no-lifetime psychiatric disorder and no family history of schizophrenia).

Instructions were:

“Pairs of words will appear on the computer screen.

dog cat

A rating scale will appear below each word pair.

dog cat

1 2 3 4 5 6

not related somewhat moderately related highly very highly

at all related related related related

Think about the word pair and then decide if it is related or unrelated in meaning.

Next, press a key from 1 – 6 on the computer keyboard to show how much you think the word pair is

related.”

Mean ratings by all 17 individuals (diagnoses combined) for the 180 semantically associated word

pairs ranged between ‘moderately related’ and ‘highly related’ (mean/sd = 4.2/0.58; range: 3.3 – 5.6);

mean ratings for the 180 semantically unassociated word pairs ranged between ‘not related at all’ and

‘somewhat related’ (mean/sd = 1.5/0.35; range: 1.2 – 2.5). Relatedness ratings for associated and

unassociated word pairs differed significantly (F1,16 = 251.87, p < .001). Mean ratings by normal

controls only (n=9) for the 180 semantically associated word pairs ranged between ‘moderately

related’ and ‘related’ (mean/sd = 4.3/0.41; range: 3.5 – 4.9); mean ratings for the 180 semantically

unassociated word pairs ranged between ‘not related at all’ and ‘somewhat related’ (mean/sd =

1.6/0.44; range: 1.2 – 2.5). Controls’ relatedness ratings for associated and unassociated word pairs

also differed significantly (F1,8 = 214.93, p < .001).

Additional Constraints on Word Stimuli: Selection of word stimuli additionally included the

following exclusion criteria: words < 3 and > 12 letters in length; orthographic similarity of prime and

target (e.g., affluence-influence); homographs (rake-leaves; abandon-leave); compound words that

could inflate the influence of word frequency in unknown but non-random ways (e.g.: ‘highway’ =

156 occurrences/4.5 million, but ‘high’ ≥ 1000 occurrences and ‘way’ ≥ 1000 occurrences). Finally,

an important goal was to include a broad range of words from the lexicon. Word length (number of



letters per word) was therefore allowed to vary within the range constrained by the minima-maxima

criterion, which provided the desired broad range of words for selection purposes.

Letter Probe Stimuli. Letter probes for the incidental priming behavioral task were generated and

ordered randomly; selection and order were therefore independent of word pair condition and

presence/absence in the words presented, resulting in 35% of the letter probes present in their

respective word pairs.



Procedure

Study participants were tested in one session that lasted approximately two hours, including

electrode application, task completion, and electrode removal. Following vision screening and

electrode application, a set of practice trials (n=5) immediately preceded the critical test trials (n=360).

Each trial sequence consisted of a prime word followed by the target word followed by the letter probe

task, which was similar to the behavior task used previously by Kutas and Hillyard (1989). Stimuli

appeared in lowercase letters (white letters on black screen) and were centered on a computer screen.

Subjects were seated 55 cm from the screen (eye-to-screen) with their chins in a fixed position. The

height of words was 0.63°, and the width of words varied from 1.3° to 5.3°. Background illumination

of the stimulus field (blank screen) was 0.4 foot candles; when single words appeared on the screen,

the luminance varied from 0.3 to 1.3 cd/m2 depending on number and content of letters. Word pairs

were presented in blocks that included trials representing each experimental condition (7 blocks of 48

trials per block; 1 block of 24 trials), with block assignment and order of within-block presentation

randomized. A 2-minute rest interval occurred between each block. Each word was presented only

once, and each target was preceded by its prime. Each letter probe followed its word pair, and subjects

were instructed to decide if the letter was present in either of the words in the pair. Subjects were not

informed that words would vary in their degree of semantic association or frequency of occurrence in

the lexicon. Subjects were instructed to be as accurate as possible, and were allowed up to 3 s to make

their response. Each trial began with the presentation of the prime for 200 ms followed by an inter-

stimulus interval (ISI) of 800 ms during which only a blank screen appeared. The target, which

immediately followed the prime-target ISI, remained on the screen for 200 ms and was followed by a

post-target ISI of 1000 ms after which the letter probe was presented; for example: admire (ISI) amuse

(ISI) r? The letter probe remained on the screen until the inter-trial interval (ITI) of 1800 ms was

triggered by the subject’s response or until 3 s had elapsed, whichever occurred first.

Electroencephalogram (EEG) Recording and Analysis. The visual EEG was recorded from 24

sites (nosetip electrode served as the reference; forehead electrode served as ground) using Ag/AgCl

electrodes inserted in a cap (Physiometrix, Inc.). Electrode placements included midlines (Fz, Cz, Pz,

Oz), laterals (F3/4, C3/4, P3/4, O1/2), temporals (T3/4, T5/6), frontal sites (Fp1/2, F7/8), and mastoids

(A1/2). Horizontal and vertical electro-oculogram (EOG) was recorded from electrodes placed at the

outer canthi of both eyes, and above and below the right eye. Channels were amplified x 10,000 K

with a Sensorium EPA-6 amplifier (bandwidth = .02 to 100 Hz). Impedances were < 30 kOhm. The



EEG was digitized at 250 samples/s (4 ms) using EEGSYS software. EEG sampling began 200 ms

prior to the onset of each prime stimulus and continued for 1000 ms post-target stimulus, which

provided an analysis epoch of 2400 ms per word-pair trial. Eye movement artifacts were corrected

using software developed by Miller et al. (1988) for the Gratton et al. procedure (1983) and adapted in

the Pittsburgh laboratory for use with the EEGSYS software. Trials including artifacts > 500 μV were

immediately rejected before vertical EOG correction; additional trials were rejected before horizontal

EOG correction if any remaining artifacts > 150 μV. In addition, every trial was visually inspected by

R.C. and S.R.S., who were blind to identity and group membership, and any trials containing

uncorrected artifacts were removed and excluded from further analyses. Averaged waveform

responses were computed from artifact-free single trials.

Mean area integration measures were computed using signed deviations from the baseline

amplitude (median amplitude for the 200-ms pre-word stimulus sample) at all electrode sites for both

primes and targets under each experimental condition. Successive latency or time windows of interest

were set based on prior work regarding visual semantic priming, and the primary focus is the N400

component measured in the time window from 300 – 500 ms post-target word stimulus.

Minima for inclusion in the analyses were ≥ 19 artifact-free event-related potential trials for each

of the eight psycholinguistic conditions (2 levels of semantic association and 4 levels of word

frequency). The initial analyses were based on the mean integrated N400 amplitude from trials in

which a behavioral response for the letter probe task occurred within 3 seconds. The mixed within-

subjects factorial design ANOVA included the between-subjects factor of diagnosis (schizophrenia, no

lifetime diagnosis of psychiatric disorder), and the repeated or within-subjects factors of semantic

association (associated, unassociated) and word frequency (very low, low, high, very high). Separate

analyses were conducted for each electrode chain (midline, lateral, and temporal), which were selected

based on prior studies of linguistic processing (Kutas and Hillyard 1989; Van Petten and Kutas 1990;

Van Petten et al. 1991), and site was entered as an additional repeated measure. Only significant F

ratios are reported. Greenhouse-Geisser correction was applied where appropriate, and corrected

degrees of freedom are reported for the associated F-ratios. The standard significance level of α = .05

was used for overall ANOVAs, and theoretically meaningful comparisons followed findings of

significant results based on the overall ANOVAs.

As an additional analytic strategy, time windows of significant differences were determined for the

contrasts of theoretical interest by using statistical tests at each sampling point along the ERP



waveforms recorded at each electrode. To control Type I error for this large number of tests, Guthrie

and Buchwald’s (1991) technique was also used, as in our previous publication (Condray et al., 2003).

This technique allows segments of waveforms that differ significantly to be identified empirically

without a priori specification of those regions, while controlling for Type I error. Monte Carlo

simulations were used to estimate the number of consecutive significant differences long enough to be

judged to not have occurred by chance with p < .05 given the temporal autocorrelation of the data.

Contiguous sample-by-sample tests are considered replications. In the present application of this

technique, t tests were performed at each point along the difference waveform for the 1 s post-target

word epoch. A minimum number of significant consecutive t tests to consider a region-wise difference

as significant was derived by simulations that accounted for the temporal autocorrelation for each

difference waveform (rxx=.97 for both condition-related and difference waveforms across electrodes

(std(rxx)<.01), and the assumption of five principal component analysis factors associated with the task

signal. For each test, 3 runs of 1000 simulations were run to create distributions from which the

number of consecutive t tests, significant at p < .05 in less than 5% of the cases was extracted, taking

the least conservative estimate; nearly identical estimates were produced for all runs. This procedure

thus controlled family-wise α at p < .05. For each semantic condition, 188 milliseconds (47 samples)

of consecutive significant differences at p<0.1 were required for the waveforms to unassociated and

associated target words to be considered statistically significantly different at p<.05 using paired t-

tests. Similarly, 196 milliseconds (49 samples) of consecutive significant differences at p<0.1 were

required for the waveforms to each word frequency condition between the two groups to be considered

statistically significantly different at p<.05 using independent samples t-tests.



Results

ERP Data

The number of artifact-free ERP trials did not differ (all p-values = n.s.) between patients and

controls (mean/sd = 298.4/50.3 and 301.9/42.5, respectively), or between the different levels of

semantic association (associated = 149.6/23.2; unassociated = 150.5/22.9) and word frequency (very

low = 75.6/12.9; low = 75.1/10.8; high = 74.7/12.5; very high = 74.8/11.2). The two groups also did

not differ under the different combinations of semantic association and word frequency (mean/sd):

associated pairs – very low = 37.2/6.5, low = 37.9/5.6, high = 37.2/6.3, very high = 37.3/5.9;

unassociated pairs – very low = 38.4/6.6, low = 37.2/5.6, high = 37.5/6.7, very high = 37.5/5.7.

N400 component

Table 3 presents the mean integrated N400 amplitude (300 – 500 ms post-target word) elicited at

lateral electrodes, collapsed across hemisphere, to associated and unassociated target words under each

word frequency condition for each diagnostic group. Results of analyses are reported below and

organized according to the psycholinguistic variables of interest. Because the task stimuli were

independent of the psycholinguistic factors of interest, results are reported first for N400 elicited

during all artifact-free ERP trials regardless of behavior response accuracy, and followed by results for

N400 elicited during correct-response trials only. Figures 1 – 6 present the grand mean ERP

waveforms for each diagnostic group under each semantic association and word frequency condition.

Diagnosis. Schizophrenia patients and normal controls did not differ for overall amplitude of

N400 recorded at any of the electrode chains (all p-values for the main effect of diagnosis = n.s.).

Semantic Association. Figures 1 and 2 present the grand mean ERP waveforms for each semantic

condition for schizophrenia patients and normal controls, respectively. The overall ANOVA for N400

recorded at midline electrodes showed a significant semantic association x electrode x diagnosis

interaction (F 2, 58 = 5.53, p = .005). Analyses of the simple effects of diagnosis indicated the source of

this interaction to be the significant effect of semantic association (N400 priming effect) for controls (F

1,13 = 4.85, p = .046), which, although greater in magnitude over posterior scalp, did not reach

statistical significance for scalp distribution (semantic association x electrode interaction: p = .053).

Patients’ N400 did not differ significantly between associated and unassociated words at midline sites

(semantic association main effect: p = .50; semantic association x midlines interaction: p = .063). For

N400 recorded at lateral electrodes, patients and controls differed at the different levels of semantic

association and scalp location (semantic association x electrode x diagnosis interaction: F 2, 40 = 6.93,



p = .005), with controls’ N400 priming effect over posterior scalp the likely source of this group

difference (semantic association x electrode interaction: F 2, 20 = 6.19, p = .012). Patients’ N400 did

not discriminate between levels of semantic relatedness at lateral sites (p = .12). N400 at temporal

electrodes did not discriminate between semantic conditions, nor did patients and controls differ in

their N400 response to semantic relationship measured at these scalp loci.

Word Frequency. Figures 3 – 6 present the grand mean ERP waveforms for each diagnostic group

under each word frequency condition. N400 recorded at lateral electrode sites revealed a significant

difference between schizophrenia patients and normal controls under the different levels of word

frequency (diagnosis x word frequency interaction: F 3, 65 = 3.51, p = .026). Simple effects analyses

indicated the locus of this interaction to be controls’ N400 discrimination between word frequency

levels (F 2, 24 = 3.75, p = .043; quadratic trend: F 1,13 = 6.31, p = .026). In contrast, patients did not

show N400 differences between word frequency levels (p = .49). A significant difference between

patients’ and controls’ N400 response to word frequency was also observed at temporal electrode sites,

(F 3, 72 = 3.41, p = .025), with their N400 response to very high frequency words the likely source of

this effect (patients = – 2.02 µV and controls = + .147 µV: F 1, 26 = 4.16, p = .052). (See Figure 7,

Supporting Information, which illustrates the results of these analyses to determine the simple effects

of diagnosis on N400 to word frequency recorded at lateral and temporal chains.) The effect of

diagnosis on the N400 response to word frequency did not reach statistical significance (p= .066) at

midline electrodes.

Modulation of semantic association by word frequency. The N400 priming effect differed only

marginally at the different levels of word frequency and only at midline electrodes (semantic

association x word frequency x site interaction: F 4, 110 = 2.39, p = .051). Moreover, patients and

controls did not differ for their N400 priming effect under the different word frequency levels at any of

the electrode chains examined (all p-values = n.s.).

Time windows of significant differences along ERP difference waveforms

Results of analyses to identify time windows of significant differences for the ERP contrasts of

interest showed effects across the 1 s post-target word recording epoch. Because these effects provide

an informative picture of the time course for linguistic access of semantic memory in schizophrenia,

results are reported and shown graphically for the entire recording epoch following target word onset.

The ERP showed both condition- and group-related variation. Results for these analyses are reported

below and organized according to the psycholinguistic factors of interest.



Semantic Association. Figures 1 and 2 show the waveforms elicited by associated and

unassociated target words for schizophrenia patients and normal controls, respectively. Table 4

presents the time windows of consecutive significant differences along the difference waveforms for

semantic condition (unassociated minus associated words) for each group. Patients’ ERP response

differed between unassociated and associated words only at Fp1 and within two time windows: 280 to

360 ms and 450 to 570 ms post-target onset. Although their ERP discrimination is of interest because

it conformed to the atypical pattern of negative priming (increased negativity to associated words), the

length or number of consecutive differences along the difference waveform did not meet the

simulations-derived criterion (188 ms/47 samples) for statistical significance. In contrast, controls

showed the typical ERP priming pattern involving enhanced negativity to unassociated words, which

occurred at multiple time windows and over diverse regions, with onsets beginning within the N400

window and later windows that extended through the end of the epoch. Most of the consecutive

differences along the difference waveform that met the simulations-derived criterion (188 ms/47

samples) for statistical significance had an anterior and bilateral scalp distribution.

The group difference waveforms for each word-pair type (associated and unassociated) are

provided in Supporting Information in the online version of this paper (Figures 8 and 9).

Word Frequency. Figures 3 – 6 show the waveforms for patient and control groups under each

word frequency condition. Table 5 presents the time windows of consecutive significant differences

along the group difference waveforms for target words under each word frequency and electrode

condition. Schizophrenia patients and normal controls differed significantly in their ERP response

under all word frequency levels, which included the following general patterns: For very low

frequency words (Figure 3), patients and controls differed within the N400 time window and across

multiple windows beginning at 600 ms and later. The length or number of consecutive differences

along the group difference waveform that met the simulations-derived criterion (196 ms/49 samples)

for statistical significance appeared over bilateral anterior scalp. For low frequency words (Figure 4),

group differences were observed primarily within the N400 time window over posterior scalp, but the

length (number) of consecutive differences along the group difference waveform did not reach the

criterion (196 ms/49 samples) for statistical significance. For high frequency words (Figure 5), the

length (number) of consecutive differences along the group difference waveform that met the criterion

(196 ms/49 samples) for statistical significance began at 500 ms post-target onset and extended

through the remaining recording epoch, and these windows were widely distributed over bilateral



posterior scalp. Significant consecutive differences along the group difference waveform for very high

frequency words (Figure 6) also began after 500 ms post-target onset, with the locations of these group

effects occurring over left posterior scalp. In summary, patients’ increased ERP negativity over

bilateral posterior scalp for high frequency words and over left posterior scalp for very high frequency

words is of interest with respect to study hypotheses regarding semantic access in schizophrenia.

N400 component elicited during correct response trials

Mean integrated N400 amplitude (300 – 500 ms post-target word) was also examined for trials on

which a correct response occurred to the letter probe task. As noted, this task was independent of the

psycholinguistic conditions of semantic meaning and word frequency, and, as reflected in the behavior

data (Table 6), was difficult for both groups resulting in an overall mean accuracy rate of 53.6%.

These analyses are therefore based on fewer artifact-free ERP trials per condition than for the analyses

based on all artifact-free ERP trials regardless of accuracy, and results must be viewed accordingly.

The number of artifact-free correct-response ERP trials did not differ (all p-values = n.s.) between

patients and controls (mean/sd = 162.4/29.4 and 169.6/24.7, respectively), or between the two groups

under the different levels of semantic association (associated = 87.5/15.1; unassociated = 78.6/12.8)

and word frequency (very low = 39.0/7.1; low = 39.9/6.0; high = 42.6/8.0; very high = 44.5/8.0), or

between patients and controls under the different combinations of semantic association and word

frequency (associated word pairs at each word frequency: very low = 20.9/4.2; low = 21.4/3.4; high =

20.7/3.9; very high = 24.5/5.2; unassociated pairs at each word frequency: very low = 18.2/3.5; low =

18.5/3.2; high = 21.9/4.8; very high = 19.9/3.5).

Schizophrenia patients and normal controls did not differ for overall amplitude of N400 elicited

during correct-response trials at any of the electrode chains (all p-values for the main effect of

diagnosis = n.s.). The principal findings for this set of analyses involved the modulation of the N400

response by the combined effects of semantic association and word frequency. For N400 elicited

during correct-response trials, the discrimination of semantic relatedness differed between diagnostic

groups at the different levels of word frequency for both lateral (semantic association x word

frequency x electrode x diagnosis interaction: F 5, 118 = 2.60, p = .03) and temporal (semantic

association x word frequency x electrode x diagnosis interaction: F 3, 66 = 3.08, p = .04) chains. The

source of the interaction effect at lateral sites is likely patients’ increased N400 negativity over

posterior scalp to very high frequency unassociated words (F 4, 47 = 2.96, p = .03), which was not

observed for controls’ N400 response (p = .35). The source of the interaction effect at temporal sites is



likely the difference in magnitude of patients’ N400 priming response under the different levels of

word frequency and scalp sites (F 3, 33 = 3.28, p = .04), which was not observed for controls (p = .51).

Although the comparisons of patients’ N400 priming response at each word frequency level failed to

reveal any statistically significant effects, the magnitude of their N400 discrimination between

semantic conditions appeared greatest for low frequency words at posterior temporal sites

(Unassociated minus Associated words = – 1.63 µV; F 1,13 = 4.44, p= .055). No significant effects

were observed for N400 elicited during correct-response trials recorded at midline electrodes (all p-

values = n.s.).

Effect on N400 of atypical antipsychotics with affinity binding for muscarinic receptors

An additional set of analyses was conducted that included only those patients receiving atypical

antipsychotic drugs for which the pharmacodynamics indicate appreciable affinity binding for

muscarinic receptors (Chew et al., 2006; Chengappa et al., 2000). Nine (56% male) of the 14

schizophrenia patients were receiving medications of this type, including clozapine or olanzapine, and

the mean dosage (chlorpromazine equivalents: Woods 2003) was 361.1 mg daily (median = 400; sd =

± 60.1; range: 300 – 450). Results of the within-group repeated measures ANOVAs based on mean

integrated N400 amplitude (300 – 500 ms post-target) recorded at midline, lateral, and temporal chains

are summarized in the following and presented graphically in Figure 10 (Supporting Information). The

only significant effect observed for this subgroup of patients was their N400 response to semantic

relationship recorded at midline electrodes. The overall ANOVA for patients’ N400 amplitude at

midline electrodes showed a significant semantic association x electrode interaction (F 2, 13 = 5.85, p =

.02), with the likely source of this effect patients’ N400 discrimination between unassociated and

associated words over anterior scalp (mean N400 priming effect at Fz and Cz = – 0.58 and – 0.97,

respectively) compared to the magnitude of their N400 difference between unassociated and associated

words over posterior sites (mean N400 priming effect at Pz and Oz = – 0.21 and – 0.06, respectively).

No other effects were observed for the semantic association factor. Moreover, N400 in this sub-group

of patients did not discriminate between the different levels of word frequency at any of the electrode

chains. Finally, N400 also did not differ under the various combinations of the semantic association

and word frequency factors at any of the electrode chains (p-values = n.s. for all semantic association x

word frequency interactions).



Behavioral Data

Table 6 shows accuracy rates, false alarms, and decision reaction times for the letter probe task for

each diagnostic group under each semantic association and word frequency condition. Results based

on the mixed within-subjects factorial design ANOVAs are reported below for each behavioral

measure.

Response accuracy (number correct). Results revealed significant main effects for semantic

association (F 1, 26 = 74.30, p < .001) and word frequency (F 3, 73 = 33.31, p < .001; linear trend: F 1, 26

= 107.34, p < .001), which indicated increased accuracy rates for associated word pairs (semantic

priming effect) and high frequency words (word frequency effect). The interaction of semantic

association and word frequency (F 2, 59 = 19.68, p < .001) showed significant semantic priming under

each word frequency level (all p-values < .001) except for high frequency words (p = .053). Overall

accuracy differed between the two groups (controls = 55%; patients = 52%: F 1, 26 = 4.49, p = .044).

Moreover, accuracy of patients and controls differed under the different levels of semantic association

(diagnosis x semantic association interaction: F 1, 26 = 5.84, p = .023) and word frequency (diagnosis x

word frequency interaction: F 3, 73 = 3.64, p = .019). The simple effects of diagnosis on semantic

association and word frequency were examined to clarify these two interaction effects. Diagnosis

influenced accuracy associated with semantic or meaning-related information, with larger semantic

priming effects observed for controls (unassociated – associated difference = – 7.4%: F 1,13 = 83.55, p

< .001) compared to patients (unassociated – associated difference = – 4.1%: F 1,13 = 15.14, p = .002).

The source of the larger priming effect for controls was their higher rate of accuracy for associated

words (controls = 59% versus patients = 54%: F 1, 26 = 6.68, p = .016). Groups did not differ for

unassociated words (controls = 51.6% versus patients = 49.9%: p = .26). The simple effects of

diagnosis on accuracy rates under the different levels of word frequency showed a significant linear

trend for word frequency for both groups (controls: F 1,13 = 132.63, p < .001; patients: F 1,13 = 23.27, p

< .001), with the likely source of this interaction being the group difference for high frequency words

(controls = 57.6% versus patients = 51.9%: F 1, 26 = 11.08, p = .003). Moreover, the accuracy

difference between the word frequency extremes (very high versus very low) was comparable for

controls and patients (9.3% versus 7%, respectively) (p = .19). Finally, patients and controls did not

differ under the joint effects of the semantic association and word frequency factors (diagnosis x

semantic association x word frequency interaction: p = .42).



False alarms (incorrect ‘Yes’ responses). Significant main effects were observed for semantic

association (F 1, 26 = 92.75, p < .001) and word frequency (F 3, 71 = 17.41, p < .001), which showed

increased rates of false alarms for unassociated words (semantic priming effect) and for very low and

low frequency words. Moreover, the interaction of semantic association and word frequency (F 3, 66 =

79.49, p < .001) involved significant semantic priming only for very low (F 1, 26 = 195.59, p < .001)

and high (F 1, 26 = 92.51, p < .001) frequency words. In contrast, negative priming (increased rate of

false alarms for associated pairs) was observed for low frequency words (F 1, 26 = 18.49, p < .001), and

absent priming (no difference between the rate of false alarms for associated and unassociated pairs)

for very high frequency words (p = .18). The overall number of false alarms did not differ between

groups (controls = 20.5%; patients = 21.6%: p = .53). However, patients and controls differed for the

number of false alarms under the different levels of semantic association (diagnosis x semantic

association interaction: F 1, 26 = 9.85, p = .004) and word frequency (diagnosis x word frequency

interaction: F 3, 71 = 4.68, p = .006). Analyses of the simple effects of diagnosis on semantic

association showed significant priming effects for both controls (unassociated – associated difference

= 7.1%: F 1,13 = 200.54, p < .001) and patients (unassociated – associated difference = 3.6%: F 1,13 =

13.22, p = .003). Analyses of the effects of diagnosis on the false alarm rate for each word-pair type

confirmed the source of this interaction to be the larger priming effect for controls; false alarm rates of

patients and controls did not differ for either associated (p = .14) or unassociated (p = .70) words.

Examination of the simple effects of diagnosis on word frequency indicated that only normal controls

showed a significant effect for the word frequency factor (patients: p = .15; controls: F 2, 25 = 27.64, p

< .001, cubic trend: F 1, 13 = 11.57, p = .005). Finally, the false alarm rates of patients and controls did

not differ under the joint effects of the semantic association and word frequency factors (diagnosis x

semantic association x word frequency interaction: p = .83)

Decision Reaction Times. No significant main effects of semantic association or word frequency

were observed for decision reaction times (harmonic means). Mean reaction times differed between

associated and unassociated words under the different levels of word frequency (semantic association x

word frequency interaction: F 3, 74 = 20.34, p < .001), with the likely source of this interaction the

negative priming effect (Unassociated RT – Associated RT = – 75.1 ms) for very low frequency words (F

1,26 = 26.89, p < .001). Typical reaction time priming effects (Unassociated RT > Associated RT) were

observed for all other word frequency levels (low = 44.4 ms; high = 35.7 ms; very high = 51.3 ms: all

p-values < .02). Overall mean reaction time did not differ between patients and controls (p = .09:



controls = 1026.9 ms versus patients = 1189.9 ms), nor did the two groups differ for the magnitude of

the semantic priming effect (p = .70: patients = 17.29 ms and controls = 10.85 ms). However,

response times differed between groups under the different levels of word frequency (diagnosis x word

frequency interaction: F 2, 62 = 3.25, p = .037), with controls showing a significant word frequency

effect (F 2, 29 = 11.97, p < .001; linear trend: F 1,13 = 19.21, p = .001) but patients’ response times not

discriminating among word frequency levels (p = .68). No other interactions involving the diagnosis

factor reached statistical significance.



Discussion

Overview: The N400 component was elicited using an incidental semantic priming paradigm, in

which semantic relatedness and word frequency were varied, and a letter probe recognition task that

was resource demanding, independent of the psycholinguistic factors of interest, and biased toward an

orthographic or shallow level of processing. Due to the task-induced bias for orthographic processing,

the N400 priming effects elicited by this paradigm are assumed to reflect primarily automatic

activation. N400 was compared between schizophrenia patients receiving atypical antipsychotic

medications and normal controls who were group-matched to patients on age, gender, and

demographic characteristics. Compared to normal controls, schizophrenia patients showed

significantly reduced N400 discrimination of semantic relatedness and word frequency. Patients

showed very limited N400 differentiation of semantically associated and unassociated word pairs, and

the priming that was observed involved the atypical pattern of increased N400 to associated target

words (negative priming), and was restricted to left anterior scalp. In contrast, controls showed the

typical pattern of enhanced ERP negativity to unassociated target words, which occurred at multiple

time windows over diverse scalp regions. Moreover, the two diagnostic groups showed markedly

different patterns of N400 response to word frequency information. Controls exhibited significant

discrimination of word frequency, with their N400 amplitude increasing in negativity as target words

decreased in frequency of occurrence in the lexicon. In contrast, patients did not show a linear

relationship between N400 and word frequency. The groups also differed in their response to

frequently occurring target words. Patients exhibited increased negativity to high and very high

frequency target words, compared to controls. Additional group differences were observed across the

recording epoch and scalp for the ERPs elicited under each word frequency level. Compared to

patients, controls showed greater negativity within the N400 time window over fronto-central scalp

sites for the extremes of the word frequency range (very low and very high). In contrast, patients

showed increased negativity, compared to controls, beginning within the N400 time window and over

bilateral posterior scalp for low and high frequency target words, and over left posterior scalp for very

high frequency target words. Thus, group differences in ERP response varied temporally and spatially

under the different levels of word frequency.

The role of acetylcholine in learning and memory is well established (reviewed in Introduction),

and it has been suggested that the increased dopamine and acetylcholine release in prefrontal cortex

that is produced by atypical agents may account for their advantageous effects on cognition (Meltzer



2004). N400 was therefore additionally examined for a subgroup of patients who were receiving

atypical antipsychotic drugs with known affinity binding for muscarinic acetylcholine receptors

(clozapine and olanzapine). N400 priming was observed at vertex for this subgroup of patients.

However, their N400 did not discriminate the different levels of word frequency. The limited N400

priming response observed in this subgroup receiving clozapine and olanzapine is consistent with the

earlier finding for schizophrenia patients who were receiving the typical antipsychotic drug haloperidol

(Condray et al., 2003). These combined results therefore suggest that haloperidol, clozapine, and

olanzapine may produce similar effects on the N400 semantic priming response in schizophrenia

patients. The absence of even a limited advantage for patients’ N400 response to word frequency

information during treatment with clozapine or olanzapine is of clinical and theoretical interest.

Storage versus access deficit in schizophrenia. A question raised at the beginning of this report is

whether the storage-access distinction can inform our understanding about semantic memory deficit in

schizophrenia. The preponderance of data from previous studies in which the storage-access

distinction was examined for this disorder has been interpreted as support for a storage deficit

(reviewed in Introduction). That prior work was based on behavior response measures, and the present

N400 study provides information regarding activation occurring earlier in the processing stream. The

pattern of N400 responding observed in the present study (see summary in Table 7) is consistent with

disturbance to both semantic storage and access as defined by Warrington, Shallice, and colleagues

(Table 1). Patients’ failure to manifest N400 differentiation of word frequency is strongly suggestive

of a disturbance to semantic access; patients’ reduced N400 semantic priming is also consistent with a

storage disorder. The significant albeit limited N400 semantic priming observed in the subgroup

receiving atypical antipsychotics with affinity binding for muscarinic receptors is potentially

important.

The most serious of the criticisms of the Warrington et al. storage-access hypothesis has been the

lack of a fully developed theory of semantic access (Rapp and Caramazza 1993), and the connectionist

model proposed by Gotts and Plaut (2002), in which semantic access is instantiated in

neuromodulatory terms, has significantly advanced this construct. It may therefore be instructive to

consider the limited but significant N400 priming effect observed for the subgroup of patients

receiving clozapine or olanzapine from the perspective of that model. Gotts and Plaut used a neural

refractory mechanism, synaptic depression, to operationalize Warrington and colleagues’ functional

“refractoriness” hypothesis, in which the ability to use the semantic system is reduced for an



abnormally prolonged period of time following activation. The heart of the Gotts-Plaut hypothesis

regarding access/refractory impairment involves the idea that semantic access in the neurologically

intact brain is determined by central neuromodulatory systems that normally operate to enhance neural

signals that are otherwise attenuated by neural refractory processes, such as synaptic depression.

Synaptic depression normally occurs when pre-synaptic neurons fire repeatedly resulting in a

temporary attenuation in the activation and sensitivity to excitatory input of post-synaptic cells. The

degree of synaptic depression in the healthy brain appears to be a function of the influence of slow-

acting, long-lasting neuromodulatory processes. Neuromodulators, such as acetylcholine, are known

to influence the probability of neurotransmitter release at both excitatory and inhibitory synapses in

cortex and neocortex, and, subsequently, to influence the degree of synaptic depression. Thus,

acetylcholine may increase the level of activation of post-synaptic neurons as a result of its influence

on the probability of transmitter release pre-synaptically. Gotts and Plaut argued that if

neuromodulatory systems are compromised, the network will be shifted disproportionately toward

synaptic depression. Viewing the N400 responding of patients in the present study within the

framework of this set of assumptions suggests the following account: Because neuromodulators such

as acetylcholine are known to suppress neurotransmitter release, and because synaptic depression

depends on the probability of neurotransmitter release, the presence of acetylcholine is expected to

reduce the degree of synaptic depression. Atypical antipsychotic drugs are known to increase

acetylcholine in cortex. The significant albeit limited N400 priming observed for the subgroup of

patients receiving the atypical agents clozapine or olanzapine may therefore reflect a

pharmacologically-induced reduction of synaptic depression within the semantic network.

As a final consideration, schizophrenia patients’ semantic priming response was also reduced

across behavioral response measures compared to controls. Moreover, patients’ responding indicated a

reduced (number correct and false alarms) and absent (reaction times) behavior discrimination of word

frequency. In contrast, previous studies based on behavioral measures have shown word frequency

effects that were greater in magnitude for patients than those observed for controls (Rossell and David

2006; Titone and Levy 2004). It is possible that differences in type of experimental tasks (word

identification versus letter-probe recognition) and semantic memory paradigm may account for these

inconsistencies across studies. Despite these differences, however, a striking consistency appears

between the N400 data from the present study and the accuracy data reported by Titone and Levy.

Those researchers showed the source of patients’ word identification errors to be high frequency words



from high density neighborhoods. That latter finding is generally consistent with our present results

showing an increased N400 negativity to very high frequency words for patients compared to controls.

Limitations of present study. Several study limitations must be recognized. Firstly, although the

clinical groups of interest were well matched with respect to age, gender, and relevant demographic

characteristics, findings from the present study must be qualified on the basis of the modest sample

size. Secondly, the present findings based on semantic relatedness and word frequency provide strong

electrophysiological support for the presence of a deficit in semantic access for schizophrenia, as well

as for disturbance of semantic storage. However, examination of the remaining two criteria originally

proposed by Warrington and colleagues for the storage-access distinction (presentation rate, response

consistency) were not accomplished for this same set of study participants. N400 data on the same

individuals are therefore desirable for all four categories of criteria. In particular, the hypothesis that a

deficit in access is fundamental to semantic disorder in schizophrenia requires examination of patients’

response variability, which is considered the hallmark of aphasic behavior (Goodglass 1993; Kolk

2007). Finally, although important information was supplied regarding the effect of atypical

antipsychotic drugs on N400 response to semantic challenge in schizophrenia, it is important to

emphasize that patients were also receiving additional medications (mood stabilizers, antidepressants,

anxiolytics). Statements regarding the effects of atypical antipsychotic medications on N400 priming

therefore must be tempered by recognition that an atypical agents only contrast was not included.

N400 semantic priming response as an endophenotype for schizophrenina. The present study

provides support for the hypothesis that the N400 response to semantic memory challenge is a

candidate risk marker for schizophrenia. Earlier conceptualizations of heritable risk markers included

the idea of deficits in diagnosed patients that are modulated by fluctuations in internal state (e.g.,

symptom exacerbations and remissions; pharmacological agents), but that never fully resolve to

normal responding (mediating vulnerability traits: Nuechterlein et al. 1992). The endophenotype

construct was more recently developed to reflect characteristics that are close to genetic variation and

linked to heritable risk (Braff et al., 2007; Gottesman and Gould 2003). The criterion for

endophenotype status that is relevant for the present data is the requirement that the characteristic is

associated with illness, and is stable or trait-related. The cumulative findings from our Pittsburgh

laboratory for the N400 priming response measured in schizophrenia patients show consistency across

samples and hospitalization status (inpatient and outpatient), behavior tasks (lexical decision,

orthography recognition), and type of antipsychotic medication (typical and atypical agents).



Acknowledgements

This research was supported by the National Institute of Mental Health (MH50631 (RC),

MH55762 (SRS), MH64023 (MSK)), and with resources and use of facilities at Western Psychiatric

Institute and Clinic, University of Pittsburgh School of Medicine. We are grateful to the patients who

served as study participants. We are also grateful to a number of individuals who made valuable

contributions to the study, including Debra Montrose, Ph.D., and the clinical and nursing staff of

Western Psychiatric Institute and Clinic. Sincere thanks to Gregory A. Miller, Ph.D., for providing

software for correction of eye movement artifacts, and to Patricia Carpenter, Ph.D., Daniel P. van

Kammen, M.D., Ph.D., and William P. Wallace, Ph.D., for their generosity during discussions with the

first author (R.C.) regarding study design and analytic approaches. Technical support was provided by

Elisa LaDonne, M.Ed., Vince Penkrot, M.S., Carole Turocy, M.S.W., and Jean Turocy, B.S.



References

Aleman, A., Hijman, R., de Haan, E.H.F., Kahn, R.S., 1999. Memory impairment in

schizophrenia: a meta-analysis. American Journal of Psychiatry 156 1358-1366.

American Psychiatric Association, 2000. Diagnostic and Statistical Manual of Mental Disorders,

Fourth Edition, Text Revision. American Psychiatric Association, Washington, D.C.

Atri A, Sherman S, Norman KA, Kirchhoff BA, Nicolas MM, Greicius MD, Cramer SC, Breiter

HC, Hasselmo ME, Stern CE (2004): Blockade of central cholinergic receptors impairs new learning

and increases proactive interference in a word paired-associate memory task. Behavioral Neuroscience

118:223-236.

Battig WF, Montague WE (1969): Category norms for verbal items in 56 categories: a replication

and extension of the Connecticut category norms. Journal of Experimental Psychology Monograph

80:1-45.

Braff DL, Freedman R, Schork NJ, Gottesman II (2007): Deconstructing schizophrenia: An

overview of the use of endophenotypes in order to understand a complex disorder. Schizophrenia

Bulletin 33:21-32.

Broks P, Preston BC, Traub M, Poppleton P, Ward C, Stahl SM (1988): Modelling dementia:

effects of scopolamine on memory and attention. Neuropsychologia 26:685-700.

Cannon, M., Caspi, A., Moffitt, T.E., Harrington, H.L., Taylor, A., Murray, R., Poulton, R., 2002.

Evidence for early-childhood, pan-developmental impairment specific to schizophreniform disorder:

Results from a longitudinal birth cohort. Archives of General Psychiatry 59 449-456.

Chengappa KNR, Pollock BG, Parepally H, Levine J, Kirshner MA, Brar JS, Zoretich RA (2000):

Anticholinergic differences among patients receiving standard clinical doses of olanzapine or

clozapine. Journal of Clinical Psychopharmacology 20:311-316.

Chew ML, Mulsant BH, Pollock BG, Lehman ME, Greenspan A, Kirshner MA, Bies RR, Kapur

S, Gharabawi G (2006): A model of anticholinergic activity of atypical antipsychotic medications.

Schizophrenia Research 88:63-72.

Condray R., 2005. Language disorder in schizophrenia as a developmental learning disorder.

Schizophrenia Research 73 5-20.

Condray R., Siegle G.J., Cohen J.D., van Kammen D.P., Steinhauer S.R., 2003. Automatic

activation of the semantic network in schizophrenia: Evidence from event-related potentials.

Biological Psychiatary, 54 1134-1148.



Condray R., Steinhauer S.R., Cohen J.D., van Kammen D.P., Kasparek A., 1999. Modulation of

language processing in schizophrenia: Effects of context and haloperidol on the event-related

potential. Biological Psychiatry 45 1336-1355.

Crow, T.J., Done, D.J., Sacker, A., 1995. Childhood precursors of psychosis as clues to its

evolutionary origins. European Archives of Psychiatry and Clinical Neuroscience 245 61-69.

Crook JM, Tomaskovic-Crook E, Copolov DL, Dean B (2000): Decreased muscarinic receptor

binding in subjects with schizophrenia: a study of the human hippocampal formation. Biological

Psychiatry 48:381-388.

Crook JM, Tomaskovic-Crook E, Copolov DL, Dean B (2001): Low muscarinic receptor binding

in prefrontal cortex from subjects with schizophrenia: a study of Brodmann’s areas 8, 9, 10, and 46

and the effects of neuroleptic drug treatment. American Journal of Psychiatry 158:918-925.

Crutch SJ, Warrington EK (2006): Partial knowledge of abstract words in patients with cortical

degenerative conditions. Neuropsychology 20:482-489.

Dean B, McLeod M, Keriakous D, McKenzie J, Scarr E (2002): Decreased muscarinic1 receptors

in the dorsolateral prefrontal cortex of subjects with schizophrenia. Molecular Psychiatry 7:1083-

1091.

DeLisi, L.E., 2001. Speech disorder in schizophrenia: review of the literature and exploration of

its relation to the uniquely human capacity for language. Schizophrenia Bulletin 27(3) 481-496.

Deng C, Huang X-F (2005): Decreased density of muscarinic receptors in the superior temporal

gyrus in schizophrenia. Journal of Neuroscience Research 81:883-890.

Doughty OJ, Done DJ, Lawrence VA, Al-Mousawi A, Ashave K (2008): Semantic memory

impairment in schizophrenia – deficit in storage or access of knowledge? Schizophrenia Research

105:40-48.

Ellis JR, Ellis KA, Bartholomeusz CR, Harrison BJ, Wesnes KA, Erskine FF, Vitetta L, Nathan PJ

(2006): Muscarinic and nicotinic receptors synergistically modulate working memory and attention in

humans. International Journal of Neuropsychopharmacology 9:175-189.

Everitt BJ, Robbins TW (1997): Central cholinergic systems and cognition. Annual Review of

Psychology 48:649-684.

First MB, Spitzer RL, Gibbon M, Williams JBW (1995-1996): Structured Clinical Interview for

DSM-IV Axis I Disorders-Patient and Non-patient Editions. New York: Biometrics Research

Department, New York State Psychiatric Institute.



Fodor J (1983): The modularity of mind. Cambridge, MA: The MIT Press.

Ford JM, Askari N, Gabrieli JDE, Mathalon DH, Tinklenberg JR, Menon V, Yesavage J (2001):

Event-related brain potential evidence of spared knowledge in Alzheimer’s disease. Psychology and

Aging 16:161-176.

Francis WN, Kucera H (1982): Frequency analysis of English usage: Lexicon and grammar.

Boston: Houghton Mifflin Company.

Goodglass H (1993): Understanding aphasia. New York: Academic Press.

Gottesman II, Gould TD (2003): The endophenotype concept in psychiatry: Etymology and

strategic intentions. American Journal of Psychiatry 160:636-645.

Gotts SF, Plaut DC (2002): The impact of synaptic depression following brain damage: A

connectionist account of “access/refractory” and “degraded-store” semantic impairments. Cognitive,

Affective, and Behavioral Neuroscience 2:187-203.

Guthrie D, Buchwald JS (1991): Significance testing of difference potentials. Psychophysiology

28:240-244.

Hasselmo HE (2006): The role of acetylcholine in learning and memory. Current opinion in

neurobiology 16:710-715.

Hodges JR, Salmon DP, Butters N (1992): Semantic memory impairment in Alzheimer’s disease:

failure of access or degraded knowledge? Neuropsychologia 30:301-314.

Jenkins JJ (1970): The 1952 Minnesota word association norms. In: Postman L, Keppel G, Eds.,

Norms of Word Association. New York: Academic Press, pp. 1-38.

Kolk H (2007): Variability is the hallmark of aphasic behaviour: grammatical behaviour is no

exception. Brain and Language 101:99-102.

Kremen, W.S., Seidman, L.J., Pepple, J.R., Lyons, M.J., Tsuang, M.T., Faraone, S.V., 1994.

Neuropsychological risk indicators for schizophrenia: a review of family studies. Schizophrenia

Bulletin 20(1) 103-119.

Kuperberg GR, Kreher DA, Ditman T (in press): What can event-related potentials tell us about

language, and perhaps even thought, in schizophrenia? International Journal of Psychophysiology.

Kutas M, Hillyard SA (1989): An electrophysiological probe of incidental semantic association.

Journal of Cognitive Neuroscience 1:38-49.

Laws KR, Al-Uzri M, Mortimer AM (2000): Lexical knowledge degradation in schizophrenia.




Marshall GR, Cofer CN (1970): Single-word free-association norms for 328 responses from the

Connecticut cultural norms for verbal items in categories. In: Postman L, Keppel G, Eds., Norms of

Word Association. New York: Academic Press, pp. 321-360.

McKay AP, McKenna PJ, Bentham P, Mortimer AM, Holbery A, Hodges JR (1996): Semantic

memory is impaired in schizophrenia. Biological Psychiatry 39:929-937.

Meltzer HY (2004): What’s atypical about atypical antipsychotic drugs? Current Opinion in

Pharmaology 4:53-57.

Miller GA, Gratton G, Yee CM (1988): Generalized implementation of an eye movement

correction procedure. Psychophysiology 25:241-243.

Minzenberg, M.J., Ober B., Vinogradov S., 2002. Semantic priming in schizophrenia: a review

and synthesis. Journal of the International Neuropsychological Society 8 699-720.

Minzenberg MJ, Poole JH, Benton C, Vinogradov S (2004): Association of anticholinergic load

with impairment of complex attention and memory in schizophrenia. American Journal of Psychiatry

161:116-124.

Nebes RD (1989): Semantic memory in Alzheimer’s Disease. Psychological Bulletin 106:377-

394.

Nelson DL, McEvoy CL, Schreiber TA (1998): The University of South Florida word association,

rhyme, and word fragment norms. http://www.usf.edu/FreeAssociation/.

Nuechterlein K.H., Dawson, M.E., Gitlin, M., Ventura, J., Goldstein, M.J., Snyder, K.S., Yee,

C.M., Mintz, J., 1992. Developmental processes in schizophrenic disorders: longitudinal studies of

vulnerability and stress. Schizophrenia Bulletin 18 387-425.

Postman L (1970): The California norms: Association as a function of word frequency. In:

Postman L, Keppel G, Eds., Norms of Word Association. New York: Academic Press, pp. 241-320.

Powchik P, Davidson M, Haroutunian V, Gabriel SM, Purohit DP, Perl DP, Harvey PD, Davis KL

(1998): Postmortem studies in schizophrenia. Schizophrenia Bulletin 24:325-341.

Raedler TJ, Bymaster FP, Tandon R, Copolov D, Dean B (2007): Towards a muscarinic

hypothesis of schizophrenia. Molecular Psychiatry 12:232-246.

Raedler TJ, Knable MB, Jones DW, Lafargue T, Urbina R, Egan MF, Pickar D, Weinberger DR

(2000): In vivo olanzapine occupancy of muscarinic acetylcholine receptors in patients with

schizophrenia. Neuropsychopharmacology 23:56-68.


http://www.usf.edu/FreeAssociation/


Raedler TJ, Knable MB, Jones DW, Urbina RA, Egan MF, Weinberger DR (2003a): Central

muscarinic acetylcholine receptor availability in patients treated with clozapine.

Neuropsychopharmacology 28:1531-1537.

Raedler TJ, Knable MB, Jones DW, Urbina RA, Gorey JG, Lee KS, Egan MF, Coppola R,

Weinberger DR (2003b): In vivo determination of muscarinic acetylycholine receptor availability in

schizophrenia. American Journal of Psychiatry 160:118-127.

Rapp B, Caramazza A (1993): On the distinction between deficits of access and deficits of

storage: A question of theory. Cognitive Neuropsychology 10:113-141.

Rogers SL, Friedman RB (2008): The underlying mechanisms of semantic memory loss in

Alzheimer’s disease and semantic dementia. Neuropsychologia 46:12-21.

Rossell SL, David AS (2006): Are semantic deficits in schizophrenia due to problems with access

or storage? Schizophrenia Research 82:121-34.

Salmon DP, Butters N, Chan AS (1999): The deterioration of semantic memory in Alzheimer’s

disease. Canadian Journal of Experimental Psychology 53:108-117.

Schacter DL, Wagner AD, Buckner RL (2000): Memory systems of 1999. In: Tulving E, Craik

FIM (eds), The Oxford Handbook of Memory, New York: Oxford University Press, pp. 627-643.

Schwartz TJ, Federmeier KD, Van Petten C, Salmon DP, Kutas M (2003): Electrophysiological

analysis of context effects in Alzheimer’s disease. Neuropsychology 17:187-201.

Strauss ME, Reynolds KS, Jayaram G, Tune LE (1990): Effects of anticholinergic medication on

memory in schizophrenia. Schizophrenia Research 3:127-129.

Sweeney JA, Keilp JG, Haas GL, Hill J, Weiden PJ (1991): Relationships between medication

treatments and neuropsychological test performance in schizophrenia. Psychiatry Research 37:297-

308.

Titone D, Levy DL (2004): Lexical competition and spoken word identification in schizophrenia.


Thorndike EL, Lorge I (1944): The teacher’s word book of 30,000 words. New York: Columbia

University.

Tulving E (1972): Episodic and semantic memory. In E. Tulving and W. Donaldson (eds.),

Organization of memory. New York: Academic Press.

Van Petten C, Kutas M (1990): Interactions between semantic context and word frequency in

event-related brain potentials. Memory and Cognition 18:380-393.



Van Petten C, Coulson S, Rubin S, Plante E, Parks M (1999): Time course of word identification

and semantic integration in spoken language. Journal of Experimental Psychology: Learning,

Memory, and Cognition 25:394-417.

Warrington EK (1975): The selective impairment of semantic memory. The Quarterly Journal of

Experimental Psychology 27:635-657.

Warrington EK, Cipolotti L (1996): Word comprehension: The distinction between refractory

and storage impairments. Brain 119:611-625.

Warrington EK, Shallice T (1979): Semantic access dyslexia. Brain 102:43-63.

Woods SW (2003): Chlorpromazine equivalent doses for the newer atypical antipsychotics.

Journal of Clinical Psychiatry 64:663-667.

Woodward ND, Purdon SE, Meltzer HY, Zald DH (2005): A meta-analysis of

neuropsychological change to clozapine, olanzapine, quetiapine, and risperidone in schizophrenia.

International Journal of Neuropsychopharmacology 8:457-472.

Woodward ND, Purdon SE, Meltzer HY, Zald DH (2007): A meta-analysis of cognitive change

with haloperidol in clinical trials of atypical antipsychotics: Dose effects and comparison to practice

effects. Schizophrenia Research 89:211-224.

Zavitsanou K, Katsifis A, Mattner F, Huang X-F (2004): Investigation of M1/M4 muscarinic

receptors in the anterior cingulated cortex in schizophrenia, bipolar disorder, and major depression

disorder. Neuropsychopharmacology 29:619-625.

SUPPORTING INFORMATION

The following supporting information for this article is available online:

Figure 7. Effect of word frequency on mean integrated N400 (μV) in schizophrenia patients

(n=14) versus normal controls (n=14).

Figure 8. Grand mean event-related potentials elicited by associated target words for normal

control subjects (n=14) and schizophrenia patients (n=14) using the incidental semantic priming

paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted black

line). Positivity is downward. The target word was presented at time zero. Regions of significant

group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along x-axes. Topography

map reflects the probability values for differences between groups for the N400 time window (300-500

ms post-target onset) across scalp.



Figure 9. Grand mean event-related potentials elicited by unassociated target words for normal

control subjects (n=14) and schizophrenia patients (n=14) using the incidental semantic priming

paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted black


group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along x-axes. Topography

map reflects the probability values for differences between groups for the N400 time window (300-500

ms post-target onset) across scalp.

Figure 10. Mean integrated N400 (μV) elicited by semantic relationship and word frequency in

schizophrenia patients (n=9; 56% male) receiving clozapine and olanzapine (M = 361 mg/day).



Tables

Table 1: Patterns of response accuracy produced by hypothetical semantic storage versus

semantic access disorders as a function of neurological etiology. (Adapted from Warrington and

Cipolotti 1996)

Table 2: Sample Characteristics

Table 3: Mean integrated N400 amplitude (μV ± sem) at each lateral electrode (collapsed across

hemisphere) under each word frequency and semantic relationship condition in schizophrenia patients

(n=14) and normal controls (n=14).

Table 4: Time windows of consecutive significant differences along difference waveforms for

semantic context (unassociated minus associated) for target words recorded at scalp electrodes for

schizophrenia patients and normal controls (all df = 13). (* p < .1 over 188 ms/47 samples)

Table 5: Time windows of consecutive significant differences along difference waveforms

between schizophrenia patients and normal controls for target words under each word frequency

condition and scalp electrode (all df = 26). (* p < .1 over 196 ms/49 samples)

Table 6: Behavioral responding to orthographic probes during incidental semantic priming under

each word frequency and semantic relationship condition in schizophrenia patients (n=14) and normal

controls (n=14). (Mean/SD)

Table 7: Summary of findings based on the N400 component and behavior response elicited

during incidental semantic priming for schizophrenia patients (n=14).

Figures

Figure 1: Grand mean event-related potentials elicited by associated and unassociated target

words for schizophrenia patients (n=14) using the incidental semantic priming paradigm. Waveforms

represent responses for associated (solid line) and unassociated (dashed line) target words. Positivity

is downward. The target word was presented at time zero. Regions of significant differences are

highlighted by color bars along the x-axes (light gray, p < .1; dark gray, p< .05). As shown in the

figure, significant differences were present for patients at Fp1, which conform to the atypical priming

pattern of greater ERP negativity to associated words (negative priming). The topography map reflects

the probability values for the discrimination between semantic conditions for the N400 time window

(300-500 ms post-target onset) across the scalp.


words for normal controls (n=14) using the incidental semantic priming paradigm. Waveforms



represent responses for associated (solid line) and unassociated (dashed line) target words. Positivity

is downward. The target word was presented at time zero. Regions of significant differences are

highlighted by color bars along the x-axes (light gray, p < .1; dark gray, p< .05). As shown in the

figure, significant differences were widely distributed across the recording epoch and scalp for

controls, which conform to the typical priming pattern of greater ERP negativity to unassociated

words. The topography map reflects the probability values for the discrimination between semantic

conditions for the N400 time window (300-500 ms post-target onset) across the scalp.

Figure 3: Grand mean event-related potentials elicited by very low frequency target words for

schizophrenia patients (n=14) and normal controls (n=14) using the incidental semantic priming

paradigm. Waveforms represent responses for patients (solid black line) and controls (dotted black


group differences are highlighted by color bars (light gray, p< .1; dark gray, p< .05) along the x-axes.

The topography map reflects the probability values for the differences between groups for the N400

time window (300-500 ms post-target onset) across the scalp.

Figure 4: Grand mean event-related potentials elicited by low frequency target words for







Figure 5: Grand mean event-related potentials elicited by high frequency target words for







Figure 6: Grand mean event-related potentials elicited by very high frequency target words for





File name: Condray_IJP_Tables-revised-final 101109.doc 1

Table 1: Patterns of response accuracy produced by hypothesized semantic storage and semantic access disorders as a function of

neurological etiology. (Adapted from Warrington and Cipolotti 1996)

Storage disorder – degenerative etiology Access disorder – vascular etiology

(degradation of semantic representations) 1 (temporary unavailability of

stored semantic representations) 2

Presentation rate Slow = Fast Slow > Fast

Word frequency High Frequency > Low Frequency High Frequency = Low Frequency

Semantic relatedness Related = Unrelated 3 (absent priming) Related < Unrelated (negative priming)

(within-category word-picture matching)

Response consistency across trials Consistent Inconsistent

1 Degenerative conditions including Alzheimer’s dementia, Pick’s disease, and viral infections 2 Acquired dyslexias 3 Storage-disorder patients showed semantic relatedness effect for cross-category pairs only



Table 2: Sample Characteristics

Schizophrenia patients Normal Controls p-value

Mean (SD/N) Mean (SD/N)

Age 34.4 ( 6.5/14) 31.4 ( 8.2/14) .32

Education (yrs) 13.5 ( 2.3/13) 14.6 ( 1.5/14) .14

WRAT-Spelling 42.1 ( 5.9/13) 45.1 ( 5.1/14) .16

WRAT-Reading 45.1 ( 4.5/13) 50.0 ( 4.6/14) .01

WASI VIQ 89.5 (15.7/13) 111.1 (10.6/14) .001

WASI Vocabulary 41.3 (12.2/13) 57.5 ( 7.9/14) .001

WAIS-R Digit Span 8.9 ( 3.7/13) 11.4 ( 3.4/14) .09

Socio-Economic Status 1

Self 25.3 ( 8.1/10) 42.4 (10.9/14) .001

Mother 28.9 (17.6/8) 38.9 (17.9/14) .22

Father 39.1 (19.3/8) 48.4 (12.6/12) .21

f f

Handedness 2 1.0

Left 1 1

Right 11 13

Race .12

African-American 8 3

Caucasian 6 11 1 Hollinshead 1975 2 Oldfield 1971



Table 3: Mean integrated N400 amplitude (μV ± sem) at each lateral electrode (collapsed across hemisphere) under each word frequency and

semantic relationship condition in schizophrenia patients (n=14) and normal controls (n=14).

Schizophrenia patients

Associated Word Pairs

Word Frequency

Very low Low High Very high

Electrode site

F3/4 - 0.583 (0.549) - 0.329 (0.611) - 0.365 (0.674) - 1.623 (0.726)

C3/4 - 1.185 (0.637) - 0.662 (0.539) - 0.258 (0.755) - 1.850 (0.828)

P3/4 - 0.817 (0.965) - 0.622 (0.867) - 0.341 (1.013) - 1.811 (1.245)

O1/2 - 1.618 (1.039) - 0.968 (1.187) - 1.121 (1.173) - 2.407 (1.395)

Unassociated Word Pairs

Word Frequency


Electrode site

F3/4 - 1.754 (0.889) - 1.507 (0.669) - 0.672 (0.542) - 0.524 (0.632)

C3/4 - 1.711 (0.751) - 1.694 (0.801) - 0.929 (0.617) - 0.657 (0.657)

P3/4 - 1.335 (0.989) - 1.560 (1.268) - 1.074 (0.962) - 0.748 (0.902)

O1/2 - 1.017 (1.272) - 1.535 (1.281) - 0.909 (1.195) - 1.657 (1.270)



Table 3 (continued): Normal Controls

Associated Word Pairs

Word Frequency


Electrode site

F3/4 - 0.964 (0.541) - 1.506 (0.541) - 1.599 (0.506) - 0.067 (0.378)

C3/4 - 1.649 (0.531) - 1.929 (0.541) - 1.633 (0.652) - 0.387 (0.637)

P3/4 - 0.538 (0.774) - 0.988 (0.716) - 0.740 (0.977) 0.962 (0.761)

O1/2 - 0.025 (0.856) - 0.131 (0.897) - 0.362 (1.225) 1.122 (0.927)

Unassociated Word Pairs

Word Frequency


Electrode site

F3/4 - 0.846 (0.566) - 0.912 (0.522) - 1.794 (0.590) - 0.923 (0.581)

C3/4 - 1.954 (0.564) - 2.072 (0.602) - 2.190 (0.718) - 1.132 (0.503)

P3/4 - 1.085 (0.670) - 1.056 (0.791) - 1.497 (0.946) - 0.092 (0.679)

O1/2 - 1.022 (0.947) - 0.600 (1.032) - 1.495 (1.222) 0.356 (1.071)



Table 4: Time windows of consecutive significant differences along difference waveforms for semantic context (unassociated minus associated) for target words recorded at scalp electrodes for schizophrenia patients and normal controls (all df = 13). (* p < .1 over 188 ms/47 samples)

Schizophrenia patients (n=14) Electrode Time windows of significant differences t-test p-value D Effect size (ms) (µV) (Cohen’s d) Fp1 280 – 360 2.24 .04 .358 0.60

450 – 570 2.33 .04 .417 0.62

Normal Controls (n=14) Electrode Time windows of significant differences t-test p-value D Effect size (ms) (µV) (Cohen’s d) Fz 380 – 570 ∗ – 3.13 .01 – .753 – 0.84 670 – 970 ∗ – 3.31 .01 – .861 – 0.89 Cz 390 – 460 – 3.03 .01 – .90 – 0.81 490 – 570 – 2.74 .02 – .973 – 0.73 660 – 930 ∗ – 3.36 .01 – .915 – 0.90 Oz 620 – 680 – 2.44 .03 – .929 – 0.65 F3 400 – 520 – 2.28 .04 – .748 – 0.61 680 – 940 ∗ – 2.69 .02 – .952 – 0.72 C3 380 – 580 ∗ – 3.38 < .01 – .946 – 0.90 590 – 660 – 2.69 .02 – .924 – 0.72 680 – 940 ∗ – 3.23 .01 – 1.027 – 0.86 P3 700 – 840 – 3.09 .01 – .723 – 0.83 T3 700 – 930 ∗ – 2.51 .03 – .934 – 0.67 T5 700 – 800 – 2.35 .04 – .922 – 0.63 810 – 920 – 2.38 .03 – .973 – 0.64 F4 490 – 560 – 2.30 .04 – .688 – 0.61 720 – 920 ∗ – 2.99 .01 – .755 – 0.80 F8 490 – 580 – 2.22 .05 – .744 – 0.59 620 – 720 – 2.12 .05 – .93 – 0.57 C4 700 – 930 ∗ – 3.51 < .01 – .906 – 0.94



Table 5: Time windows of consecutive significant differences along difference waveforms between schizophrenia patients and normal controls for target words under each word frequency condition and scalp electrode (all df = 26). (* p < .1 over 196 ms/49 samples) Word frequency and Electrode Time windows of significant differences t-test p-value D Effect size

(ms) (µV) (Cohen’s d) Very low Word Frequency

Fz 380 – 570 – 2.51 .02 – 2.064 – 0.95 590 – 740 – 1.95 .06 – 2.407 – 0.74

920 – 1000 – 2.10 .05 – 2.505 – 0.79 Cz 390 – 580 – 2.52 .02 – 3.029 – 0.95 600 – 720 – 2.10 .05 – 2.703 – 0.79

Fp1 820 – 1000 – 1.98 .06 – 1.423 – 0.75 F3 390 – 590 ∗ – 2.31 .03 – 1.718 – 0.87

600 – 750 – 1.98 .06 – 2.299 – 0.75 860 – 1000 – 2.12 .04 – 2.431 – 0.80

F7 620 – 690 – 1.80 .08 – 1.312 – 0.68 Fp2 480 – 640 – 1.97 .06 – 1.727 – 0.75

760 – 960 ∗ – 1.98 .06 – 2.152 – 0.75 F4 390 – 560 – 2.28 .03 – 1.812 – 0.86

570 – 740 – 1.99 .06 – 2.281 – 0.75 900 – 1000 – 1.95 .06 – 2.569 – 0.74

Low Word Frequency

O1 340 – 400 2.86 .01 3.664 1.08 T5 330 – 440 2.90 .01 3.212 1.10 460 – 640 1.91 .07 3.142 0.72 760 – 840 1.87 .07 3.162 0.71 F4 80 – 150 – 2.25 .03 – .527 – 0.85



Table 5 (continued): Word frequency and Electrode Time windows of significant differences t-test p-value D Effect size (ms) (µV) (Cohen’s d) High Word Frequency Pz 780 – 880 2.11 .04 2.539 0.80 Oz 590 – 860 ∗ 2.18 .04 2.289 0.82 C3 740 – 880 2.04 .05 2.179 0.77 P3 640 – 900 ∗ 2.79 .01 2.866 1.06 O1 340 – 420 2.73 .01 3.441 1.03

530 – 1000 ∗ 3.27 < .01 3.461 1.23 T3 350 – 460 2.85 .01 1.233 1.08 540 – 910 ∗ 2.54 .02 1.786 0.96

T5 190 – 260 2.71 .01 1.069 1.03 340 – 480 3.00 .01 2.853 1.13 500 – 1000 ∗ 3.03 .01 3.286 1.14 P4 630 – 860 ∗ 2.38 .02 2.745 0.90

O2 190 – 260 2.99 .01 1.466 1.13 540 – 890 ∗ 2.87 .01 2.984 1.09 T4 180 – 280 3.28 < .01 .998 1.24 360 – 440 3.52 < .01 1.252 1.33 600 – 900 ∗ 2.68 .01 1.978 1.01 T6 190 – 270 3.29 < .01 1.457 1.25 350 – 470 3.15 < .01 3.213 1.19 540 – 890 ∗ 2.36 .03 3.274 0.89 910 – 1000 2.12 .04 2.406 0.80 Very High Word Frequency Fz 380 – 490 – 2.47 .02 – 1.897 – 0.94 Cz 410 – 480 – 2.39 .02 – 2.942 – 0.90 Oz 630 – 700 2.00 .06 2.190 0.76 P3 620 – 800 2.32 .03 2.369 0.88 O1 580 – 790 ∗ 2.31 .03 2.662 0.87 T5 590 – 800 ∗ 2.18 .04 2.944 0.82 F4 190 – 350 – 2.77 .01 – 1.188 – 1.05 380 – 490 – 2.22 .04 – 1.751 – 0.84



Table 6: Behavioral responding to orthographic probes during incidental semantic priming under each word frequency and semantic relationship condition in schizophrenia patients (n=14) and normal controls (n=14). (Mean/SD)

Schizophrenia patients

Word Frequency


Associated words

Number correct (%) .52 (.05) .54 (.07) .51 (.07) .59 (.11)

False Alarms (%) .16 (.05) .25 (.09) .18 (.07) .20 (.08)

Reaction Times (ms) 1212.5 (231.3) 1175.1 (219.7) 1163.0 (232.6) 1174.5 (189.2)

Unassociated words

Number correct (%) .46 (.06) .49 (.07) .53 (.06) .52 (.06)

False Alarms (%) .29 (.07) .19 (.06) .25 (.07) .20 (.07)


Normal Controls

Word Frequency


Associated words

Number correct (%) .57 (.06) .56 (.05) .56 (.04) .67 (.04)

False Alarms (%) .13 (.04) .25 (.04) .14 (.02) .16 (.03)


Unassociated words

Number correct (%) .46 (.04) .48 (.05) .59 (.07) .54 (.04)

False Alarms (%) .31 (.05) .22 (.03) .24 (.04) .19 (.04)




Table 7: Summary of findings based on the N400 component and behavior response elicited during incidental semantic priming for

schizophrenia patients (n=14).

Absent or Reduced Normal Exaggerated

N400 (μV)

Semantic priming effect X

Word Frequency effect X

Behavioral data

Semantic priming effect

Accuracy X

False alarm X

Reaction time X

Word Frequency effect

Accuracy X

False alarm X

Reaction time X

X = response compared to normal controls (n=14)


Figures


words for schizophrenia patients (n=14) using the incidental semantic priming paradigm.

Waveforms represent responses for associated (solid line) and unassociated (dashed line) target

words. Positivity is downward. The target word was presented at time zero. Regions of

significant differences are highlighted by color bars along the x-axes (light pink, p < .1; dark

pink, p< .05). As shown in the figure, significant differences were present for patients at Fp1,

which conform to the atypical priming pattern of greater ERP negativity to associated words

(negative priming). The topography map reflects the probability values for the discrimination

between semantic conditions for the N400 time window (300-500 ms post-target onset) across

the scalp.


words for normal controls (n=14) using the incidental semantic priming paradigm. Waveforms

represent responses for associated (solid line) and unassociated (dashed line) target words.

Positivity is downward. The target word was presented at time zero. Regions of significant

differences are highlighted by color bars along the x-axes (light pink, p < .1; dark pink, p< .05).

As shown in the figure, significant differences were widely distributed across the recording

epoch and scalp for controls, which conform to the typical priming pattern of greater ERP

negativity to unassociated words. The topography map reflects the probability values for the

discrimination between semantic conditions for the N400 time window (300-500 ms post-target

onset) across the scalp.

Figure 3: Grand mean event-related potentials elicited by very low frequency target words

for schizophrenia patients (n=14) and normal controls (n=14) using the incidental semantic

priming paradigm. Waveforms represent responses for patients (solid green line) and controls

(dotted black line). Positivity is downward. The target word was presented at time zero.

Regions of significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05)

along the x-axes. The topography map reflects the probability values for the differences between

groups for the N400 time window (300-500 ms post-target onset) across the scalp.

Figure 4: Grand mean event-related potentials elicited by low frequency target words for


paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted

File name: Condray et al Figure Legends for Color Waveforms etc 101409.doc


File name: Condray et al Figure Legends for Color Waveforms etc 101409.doc

black line). Positivity is downward. The target word was presented at time zero. Regions of

significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along the

x-axes. The topography map reflects the probability values for the differences between groups

for the N400 time window (300-500 ms post-target onset) across the scalp.

Figure 5: Grand mean event-related potentials elicited by high frequency target words for


paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted

black line). Positivity is downward. The target word was presented at time zero. Regions of

significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along the

x-axes. The topography map reflects the probability values for the differences between groups

for the N400 time window (300-500 ms post-target onset) across the scalp.

Figure 6: Grand mean event-related potentials elicited by very high frequency target words

for schizophrenia patients (n=14) and normal controls (n=14) using the incidental semantic

priming paradigm. Waveforms represent responses for patients (solid green line) and controls

(dotted black line). Positivity is downward. The target word was presented at time zero.

Regions of significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05)

along the x-axes. The topography map reflects the probability values for the differences between

groups for the N400 time window (300-500 ms post-target onset) across the scalp.

Effects of word frequency on semantic memory in schizophrenia:Electrophysiological evidence for a deficit in linguistic access

Ruth Condray, Greg J. Siegle, Matcheri Keshavan, Stuart R. Steinhauer

SUPPORTING INFORMATION

Condray et al Supporting Information IJP paper - final 100909

Figure 7: Effect of word frequency on mean integrated N400 (μV)in schizophrenia patients (n=14) versus normal controls (n=14)

N400 (μV) at Lateral Chain:Diagnosis x Word frequency (p < .03)

N400 (μV) at Temporal Chain:Diagnosis x Word frequency (p < .03)


Figure 8. Grand mean event-related potentials elicited by associated target words for normal control subjects (n=14) and schizophrenia patients (n=14) using the incidental semantic priming paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted black line). Positivity is downward. The target word was presented at time zero. Regions of significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along x-axes. Topography map reflects the probability values for differences between groups for the N400 time window (300-500 ms post-target onset) across scalp.


Figure 8: Associated target words

0

0.02

0.04

0.06

0.08

0.1

Fz

Cz

Pz

Oz

Fp1

F3F7

C3

P3

T3

T5

O1

Fp2

F4 F8

C4

P4

T4

T6

O2A1

0 0.5 1

- 202

A2

Control

Schizophrenia

300 – 500 ms


Figure 9. Grand mean event-related potentials elicited by unassociated target words for normal control subjects (n=14) and schizophrenia patients (n=14) using the incidental semantic priming paradigm. Waveforms represent responses for patients (solid green line) and controls (dotted black line). Positivity is downward. The target word was presented at time zero. Regions of significant group differences are highlighted by color bars (yellow, p< .1; red, p< .05) along x-axes. Topography map reflects the probability values for differences between groups for the N400 time window (300-500 ms post-target onset) across scalp.


Figure 9: Unassociated target words

0

0.02

0.04

0.06

0.08

0.1

300 – 500 ms

Fz

Cz

Pz

Oz

Fp1

F3F7

C3

P3

T3

T5

O1

Fp2

F4 F8

C4

P4

T4

T6

O2A1

0 0.5 1

- 2

0

2

A2

Control

Schizophrenia


Figure 10: Mean integrated N400 (μV) elicited by semantic relationship and word frequency in schizophrenia patients (n=9; 56% male) receiving clozapine or olanzapine (M=361 mg/day)

N400 priming x midlines interactionp< .05

N400 to Word Frequency (midlines) p=n.s.(also, lateral & temporal chains, p= n.s.)

Effects of word frequency on semantic memory in ... Language Issue...Condray et al. N400 in...

Documents

Transcript of Effects of word frequency on semantic memory in ... Language Issue...Condray et al. N400 in...